Best Practices for Ranking Attributes

To make the most of your ranking attributes, consider adhering to the following practices.

Use a naming convention

After you create and save a ranking attribute, it becomes available in the drop-down menus for attribute selection across the Business Manager. To be able to identify the ranking attribute you are looking for in a list of tens and hundreds of attributes, use a consistent, user-friendly naming convention.

• To distinguish between attributes from the data model and ranking attribute, you can use an rc_ prefix in the names of ranking cocktails.
• If you are creating a ranking attribute which you can reuse across multiple trigger-based rules, you can name the ranking cocktail after its ingredients. For example: rc_70_freshness_30_margin.
• If you are creating a ranking attribute which you want to use in a single trigger-based rule to achieve a specific goal, you can name the ranking cocktail after the rule or the goal. For example: rc_campaign_summer_sale or rc_summer_sale.

Test the behavior of each attribute

For ranking attribute that consist of more than two attributes, start simple:

1. Begin by combining the first two attributes
2. Test the behavior of the ranking attribute.
3. When you have verified that the blend of attributes works as expected, add the next attribute on your list
4. Test again.
5. Repeat Steps 3 and 4 until you have created the complete ranking cocktail.

This step-by-step process lets you identify issues with your ranking attributes early on.

Add at least one attribute for which data is always available

For ranking attributes that rely on data accumulated over time, add at least one attribute for which data is always available. This ensures that your ranking attribute works from the beginning.

Example

You want to create a ranking attribute which promotes new arrivals and combines the following attributes:

• Age
• Conversion rate
• Product review rating

For newly added items, the value of the ranking attribute is always 0 because no historical data is available. Any sorting or filtering based on that ranking attribute ignores the newest items in your catalog and does not show them to your shoppers. To make sure that the latest additions to your catalog appear, add another attribute for which data is always available. For example, items on stock.

Show the ranking attribute and its values in your preview environment

When you test your ranking attribute you might want to be able to see its values and how they influence the sorting or filtering of items in your catalog. To do that, you can include the ranking attribute in the attributes enabled in the Displayed Fields in the System Settings of the Merchandising Studio.

Align the ranking attribute with your eCommerce strategy and your shoppers' behaviors

Create ranking attributes that meet your eCommerce strategy. Add behavioral attributes to create ranking attributes that adjust to the behavior of your shopper.

Example

You are carrying out a summer sale. You can sort the items in the sale based on the shopper's brand affinity, the age of the item, and the conversion rates for the item. This way, each shopper will see an individual selection of new, popular items from their preferred brands at the top of the sale.

Choose normalization based on the values and the value distribution of the attribute

If all attribute values exist on a 0.0--1.0 or a 0-100 scale, choose the respective pre-normalized setting.

If all attribute values and ranges are equally important, choose linear normalization.

If some attribute values and ranges are more important than others and need to have a significantly greater influence on the ranking cocktail, choose logarithmic normalization.

The following table lists some best practice recommendations for choosing normalization, based on attribute.

Example

You want to create a ranking attribute which relies on unit sales per week, product rating, and other attributes that have equally many items for low and high values.

The following figure shows the value distribution of the "unit sales per week" attribute:

Figure 1: Sample value distribution of a "unit sales per week" attribute

If you use linear normalization, more than 90% of all items will have values smaller than 40%. In this case, the number of sales will not influence the final value of the ranking cocktail significantly. Instead, you can use logarithmic normalization to impact the ranking cocktail value more greatly.

Choose your pushing setting based on how you want to rate attribute values

Configure the pushing setting when the ranking attribute consists of attributes that require mixed sorting. For example, for one attribute, you need to promote items with lower values (sort from high to low) and for another, you need to promote items with higher values (sort from low to high).

If you want to promote items with high attribute values, preserve the default selection High values on top. For example, this setting is useful for attributes that represent the number of sales in bestseller cocktails or attributes that represent product views in popular items cocktails.

If you want to promote items with low attribute values, push Low values on top. For example, this setting is useful for attributes that represent number of days online in new arrivals cocktails or attributes that represent low stock quantities in clearance cocktails.

When all attributes require the same sorting, you can leave the interpretation of their values to the sorting setting in the rules that use them.

Pay attention to the sorting setting in rules that use the ranking attribute

When you create your ranking attribute, you need to know how to promote items based on the ranking cocktail values. For example, for some ranking cocktails, you might want to promote items with lower values. For other ranking cocktails, the reverse might be true.

When you use a ranking attribute in a rule, you can indicate to Fredhopper which ranking cocktail values to promote using the sorting option. This setting builds on your normalization and direction settings in the ranking cocktail and assumes that you have configured them correctly.

If the ranking attribute doesn't behave as expected in any rules that rely on it, make sure to examine the sorting settings in the rules.

Clean up unused ranking attributes regularly

Drop-down menus for attribute selection list all available attributes and ranking attributes, including unused ones. Clean up the ranking attributes regularly to keep your drop-down menus more compact and easier to navigate.